Overview
- On Nov. 10, Stanford professor and World Labs co‑founder Fei‑Fei Li published a long essay arguing that spatial intelligence is the decisive next step for AI.
- She says today’s large language and multimodal models handle abstract knowledge yet falter on distance, direction, object rotation, navigation, and sustained video coherence.
- Li proposes a class of world models defined by three capabilities—generative, multimodal, interactive—to represent, simulate, and update consistent 3D worlds over time.
- World Labs, founded in early 2024, shared a limited preview of an early system called Marble for selected creators, with broader availability not announced.
- The roadmap calls for a universal training objective, large multimodal datasets including depth and tactile signals, new 3D/4D‑native architectures, and improved sensing to unlock uses in robotics, creative tools, and scientific discovery.